In this paper, we address how to design a distributed movement strategy for mobile collectors, which can be either physical mobile agents or query/collector packets periodically launched by the sink, to achieve successful data gathering in wireless sensor networks. Formulating the problem as general random walks on a graph composed of sensor nodes, we analyze how much data can be successfully gathered in time under any Markovian random-walk movement strategies for mobile collectors moving over a graph (or network), while each sensor node is equipped with limited buffer space and data arrival rates are heterogeneous over different sensor nodes.
In particular, from the analysis, we obtain the optimal movement strategy among a class of Markovian strategies so as to minimize the data loss rate over all sensor nodes, and explain how such an optimal movement strategy can be made to work in adistributed fashion. We demonstrate that our distributed optimal movement strategy can lead to about 2 times smaller loss rate than a standard random walk strategy under diverse scenarios. In particular, our strategy results in up to 70% cost savings for the deployment of multiple collectors to achieve the target data loss rate than the standard random walk strategy.